2 * linux/fs/ext3/inode.c
4 * Copyright (C) 1992, 1993, 1994, 1995
5 * Remy Card (card@masi.ibp.fr)
6 * Laboratoire MASI - Institut Blaise Pascal
7 * Universite Pierre et Marie Curie (Paris VI)
11 * linux/fs/minix/inode.c
13 * Copyright (C) 1991, 1992 Linus Torvalds
15 * Goal-directed block allocation by Stephen Tweedie
16 * (sct@redhat.com), 1993, 1998
17 * Big-endian to little-endian byte-swapping/bitmaps by
18 * David S. Miller (davem@caip.rutgers.edu), 1995
19 * 64-bit file support on 64-bit platforms by Jakub Jelinek
20 * (jj@sunsite.ms.mff.cuni.cz)
22 * Assorted race fixes, rewrite of ext3_get_block() by Al Viro, 2000
25 #include <linux/module.h>
27 #include <linux/time.h>
28 #include <linux/ext3_jbd.h>
29 #include <linux/jbd.h>
30 #include <linux/smp_lock.h>
31 #include <linux/highuid.h>
32 #include <linux/pagemap.h>
33 #include <linux/quotaops.h>
34 #include <linux/string.h>
35 #include <linux/buffer_head.h>
36 #include <linux/writeback.h>
37 #include <linux/mpage.h>
38 #include <linux/uio.h>
39 #include <linux/vserver/xid.h>
44 * Test whether an inode is a fast symlink.
46 static inline int ext3_inode_is_fast_symlink(struct inode *inode)
48 int ea_blocks = EXT3_I(inode)->i_file_acl ?
49 (inode->i_sb->s_blocksize >> 9) : 0;
51 return (S_ISLNK(inode->i_mode) &&
52 inode->i_blocks - ea_blocks == 0);
55 /* The ext3 forget function must perform a revoke if we are freeing data
56 * which has been journaled. Metadata (eg. indirect blocks) must be
57 * revoked in all cases.
59 * "bh" may be NULL: a metadata block may have been freed from memory
60 * but there may still be a record of it in the journal, and that record
61 * still needs to be revoked.
64 int ext3_forget(handle_t *handle, int is_metadata,
65 struct inode *inode, struct buffer_head *bh,
72 BUFFER_TRACE(bh, "enter");
74 jbd_debug(4, "forgetting bh %p: is_metadata = %d, mode %o, "
76 bh, is_metadata, inode->i_mode,
77 test_opt(inode->i_sb, DATA_FLAGS));
79 /* Never use the revoke function if we are doing full data
80 * journaling: there is no need to, and a V1 superblock won't
81 * support it. Otherwise, only skip the revoke on un-journaled
84 if (test_opt(inode->i_sb, DATA_FLAGS) == EXT3_MOUNT_JOURNAL_DATA ||
85 (!is_metadata && !ext3_should_journal_data(inode))) {
87 BUFFER_TRACE(bh, "call journal_forget");
88 return ext3_journal_forget(handle, bh);
94 * data!=journal && (is_metadata || should_journal_data(inode))
96 BUFFER_TRACE(bh, "call ext3_journal_revoke");
97 err = ext3_journal_revoke(handle, blocknr, bh);
99 ext3_abort(inode->i_sb, __FUNCTION__,
100 "error %d when attempting revoke", err);
101 BUFFER_TRACE(bh, "exit");
106 * Work out how many blocks we need to progress with the next chunk of a
107 * truncate transaction.
110 static unsigned long blocks_for_truncate(struct inode *inode)
112 unsigned long needed;
114 needed = inode->i_blocks >> (inode->i_sb->s_blocksize_bits - 9);
116 /* Give ourselves just enough room to cope with inodes in which
117 * i_blocks is corrupt: we've seen disk corruptions in the past
118 * which resulted in random data in an inode which looked enough
119 * like a regular file for ext3 to try to delete it. Things
120 * will go a bit crazy if that happens, but at least we should
121 * try not to panic the whole kernel. */
125 /* But we need to bound the transaction so we don't overflow the
127 if (needed > EXT3_MAX_TRANS_DATA)
128 needed = EXT3_MAX_TRANS_DATA;
130 return EXT3_DATA_TRANS_BLOCKS + needed;
134 * Truncate transactions can be complex and absolutely huge. So we need to
135 * be able to restart the transaction at a conventient checkpoint to make
136 * sure we don't overflow the journal.
138 * start_transaction gets us a new handle for a truncate transaction,
139 * and extend_transaction tries to extend the existing one a bit. If
140 * extend fails, we need to propagate the failure up and restart the
141 * transaction in the top-level truncate loop. --sct
144 static handle_t *start_transaction(struct inode *inode)
148 result = ext3_journal_start(inode, blocks_for_truncate(inode));
152 ext3_std_error(inode->i_sb, PTR_ERR(result));
157 * Try to extend this transaction for the purposes of truncation.
159 * Returns 0 if we managed to create more room. If we can't create more
160 * room, and the transaction must be restarted we return 1.
162 static int try_to_extend_transaction(handle_t *handle, struct inode *inode)
164 if (handle->h_buffer_credits > EXT3_RESERVE_TRANS_BLOCKS)
166 if (!ext3_journal_extend(handle, blocks_for_truncate(inode)))
172 * Restart the transaction associated with *handle. This does a commit,
173 * so before we call here everything must be consistently dirtied against
176 static int ext3_journal_test_restart(handle_t *handle, struct inode *inode)
178 jbd_debug(2, "restarting handle %p\n", handle);
179 return ext3_journal_restart(handle, blocks_for_truncate(inode));
182 static void ext3_truncate_nocheck (struct inode *inode);
185 * Called at the last iput() if i_nlink is zero.
187 void ext3_delete_inode (struct inode * inode)
191 if (is_bad_inode(inode))
194 handle = start_transaction(inode);
195 if (IS_ERR(handle)) {
196 /* If we're going to skip the normal cleanup, we still
197 * need to make sure that the in-core orphan linked list
198 * is properly cleaned up. */
199 ext3_orphan_del(NULL, inode);
207 ext3_truncate_nocheck(inode);
209 * Kill off the orphan record which ext3_truncate created.
210 * AKPM: I think this can be inside the above `if'.
211 * Note that ext3_orphan_del() has to be able to cope with the
212 * deletion of a non-existent orphan - this is because we don't
213 * know if ext3_truncate() actually created an orphan record.
214 * (Well, we could do this if we need to, but heck - it works)
216 ext3_orphan_del(handle, inode);
217 EXT3_I(inode)->i_dtime = get_seconds();
220 * One subtle ordering requirement: if anything has gone wrong
221 * (transaction abort, IO errors, whatever), then we can still
222 * do these next steps (the fs will already have been marked as
223 * having errors), but we can't free the inode if the mark_dirty
226 if (ext3_mark_inode_dirty(handle, inode))
227 /* If that failed, just do the required in-core inode clear. */
230 ext3_free_inode(handle, inode);
231 ext3_journal_stop(handle);
234 clear_inode(inode); /* We must guarantee clearing of inode... */
237 static int ext3_alloc_block (handle_t *handle,
238 struct inode * inode, unsigned long goal, int *err)
240 unsigned long result;
242 result = ext3_new_block(handle, inode, goal, err);
250 struct buffer_head *bh;
253 static inline void add_chain(Indirect *p, struct buffer_head *bh, __le32 *v)
255 p->key = *(p->p = v);
259 static inline int verify_chain(Indirect *from, Indirect *to)
261 while (from <= to && from->key == *from->p)
267 * ext3_block_to_path - parse the block number into array of offsets
268 * @inode: inode in question (we are only interested in its superblock)
269 * @i_block: block number to be parsed
270 * @offsets: array to store the offsets in
271 * @boundary: set this non-zero if the referred-to block is likely to be
272 * followed (on disk) by an indirect block.
274 * To store the locations of file's data ext3 uses a data structure common
275 * for UNIX filesystems - tree of pointers anchored in the inode, with
276 * data blocks at leaves and indirect blocks in intermediate nodes.
277 * This function translates the block number into path in that tree -
278 * return value is the path length and @offsets[n] is the offset of
279 * pointer to (n+1)th node in the nth one. If @block is out of range
280 * (negative or too large) warning is printed and zero returned.
282 * Note: function doesn't find node addresses, so no IO is needed. All
283 * we need to know is the capacity of indirect blocks (taken from the
288 * Portability note: the last comparison (check that we fit into triple
289 * indirect block) is spelled differently, because otherwise on an
290 * architecture with 32-bit longs and 8Kb pages we might get into trouble
291 * if our filesystem had 8Kb blocks. We might use long long, but that would
292 * kill us on x86. Oh, well, at least the sign propagation does not matter -
293 * i_block would have to be negative in the very beginning, so we would not
297 static int ext3_block_to_path(struct inode *inode,
298 long i_block, int offsets[4], int *boundary)
300 int ptrs = EXT3_ADDR_PER_BLOCK(inode->i_sb);
301 int ptrs_bits = EXT3_ADDR_PER_BLOCK_BITS(inode->i_sb);
302 const long direct_blocks = EXT3_NDIR_BLOCKS,
303 indirect_blocks = ptrs,
304 double_blocks = (1 << (ptrs_bits * 2));
309 ext3_warning (inode->i_sb, "ext3_block_to_path", "block < 0");
310 } else if (i_block < direct_blocks) {
311 offsets[n++] = i_block;
312 final = direct_blocks;
313 } else if ( (i_block -= direct_blocks) < indirect_blocks) {
314 offsets[n++] = EXT3_IND_BLOCK;
315 offsets[n++] = i_block;
317 } else if ((i_block -= indirect_blocks) < double_blocks) {
318 offsets[n++] = EXT3_DIND_BLOCK;
319 offsets[n++] = i_block >> ptrs_bits;
320 offsets[n++] = i_block & (ptrs - 1);
322 } else if (((i_block -= double_blocks) >> (ptrs_bits * 2)) < ptrs) {
323 offsets[n++] = EXT3_TIND_BLOCK;
324 offsets[n++] = i_block >> (ptrs_bits * 2);
325 offsets[n++] = (i_block >> ptrs_bits) & (ptrs - 1);
326 offsets[n++] = i_block & (ptrs - 1);
329 ext3_warning (inode->i_sb, "ext3_block_to_path", "block > big");
332 *boundary = (i_block & (ptrs - 1)) == (final - 1);
337 * ext3_get_branch - read the chain of indirect blocks leading to data
338 * @inode: inode in question
339 * @depth: depth of the chain (1 - direct pointer, etc.)
340 * @offsets: offsets of pointers in inode/indirect blocks
341 * @chain: place to store the result
342 * @err: here we store the error value
344 * Function fills the array of triples <key, p, bh> and returns %NULL
345 * if everything went OK or the pointer to the last filled triple
346 * (incomplete one) otherwise. Upon the return chain[i].key contains
347 * the number of (i+1)-th block in the chain (as it is stored in memory,
348 * i.e. little-endian 32-bit), chain[i].p contains the address of that
349 * number (it points into struct inode for i==0 and into the bh->b_data
350 * for i>0) and chain[i].bh points to the buffer_head of i-th indirect
351 * block for i>0 and NULL for i==0. In other words, it holds the block
352 * numbers of the chain, addresses they were taken from (and where we can
353 * verify that chain did not change) and buffer_heads hosting these
356 * Function stops when it stumbles upon zero pointer (absent block)
357 * (pointer to last triple returned, *@err == 0)
358 * or when it gets an IO error reading an indirect block
359 * (ditto, *@err == -EIO)
360 * or when it notices that chain had been changed while it was reading
361 * (ditto, *@err == -EAGAIN)
362 * or when it reads all @depth-1 indirect blocks successfully and finds
363 * the whole chain, all way to the data (returns %NULL, *err == 0).
365 static Indirect *ext3_get_branch(struct inode *inode, int depth, int *offsets,
366 Indirect chain[4], int *err)
368 struct super_block *sb = inode->i_sb;
370 struct buffer_head *bh;
373 /* i_data is not going away, no lock needed */
374 add_chain (chain, NULL, EXT3_I(inode)->i_data + *offsets);
378 bh = sb_bread(sb, le32_to_cpu(p->key));
381 /* Reader: pointers */
382 if (!verify_chain(chain, p))
384 add_chain(++p, bh, (__le32*)bh->b_data + *++offsets);
402 * ext3_find_near - find a place for allocation with sufficient locality
404 * @ind: descriptor of indirect block.
406 * This function returns the prefered place for block allocation.
407 * It is used when heuristic for sequential allocation fails.
409 * + if there is a block to the left of our position - allocate near it.
410 * + if pointer will live in indirect block - allocate near that block.
411 * + if pointer will live in inode - allocate in the same
414 * In the latter case we colour the starting block by the callers PID to
415 * prevent it from clashing with concurrent allocations for a different inode
416 * in the same block group. The PID is used here so that functionally related
417 * files will be close-by on-disk.
419 * Caller must make sure that @ind is valid and will stay that way.
422 static unsigned long ext3_find_near(struct inode *inode, Indirect *ind)
424 struct ext3_inode_info *ei = EXT3_I(inode);
425 __le32 *start = ind->bh ? (__le32*) ind->bh->b_data : ei->i_data;
427 unsigned long bg_start;
428 unsigned long colour;
430 /* Try to find previous block */
431 for (p = ind->p - 1; p >= start; p--)
433 return le32_to_cpu(*p);
435 /* No such thing, so let's try location of indirect block */
437 return ind->bh->b_blocknr;
440 * It is going to be refered from inode itself? OK, just put it into
441 * the same cylinder group then.
443 bg_start = (ei->i_block_group * EXT3_BLOCKS_PER_GROUP(inode->i_sb)) +
444 le32_to_cpu(EXT3_SB(inode->i_sb)->s_es->s_first_data_block);
445 colour = (current->pid % 16) *
446 (EXT3_BLOCKS_PER_GROUP(inode->i_sb) / 16);
447 return bg_start + colour;
451 * ext3_find_goal - find a prefered place for allocation.
453 * @block: block we want
454 * @chain: chain of indirect blocks
455 * @partial: pointer to the last triple within a chain
456 * @goal: place to store the result.
458 * Normally this function find the prefered place for block allocation,
459 * stores it in *@goal and returns zero. If the branch had been changed
460 * under us we return -EAGAIN.
463 static int ext3_find_goal(struct inode *inode, long block, Indirect chain[4],
464 Indirect *partial, unsigned long *goal)
466 struct ext3_inode_info *ei = EXT3_I(inode);
467 /* Writer: ->i_next_alloc* */
468 if (block == ei->i_next_alloc_block + 1) {
469 ei->i_next_alloc_block++;
470 ei->i_next_alloc_goal++;
473 /* Reader: pointers, ->i_next_alloc* */
474 if (verify_chain(chain, partial)) {
476 * try the heuristic for sequential allocation,
477 * failing that at least try to get decent locality.
479 if (block == ei->i_next_alloc_block)
480 *goal = ei->i_next_alloc_goal;
482 *goal = ext3_find_near(inode, partial);
490 * ext3_alloc_branch - allocate and set up a chain of blocks.
492 * @num: depth of the chain (number of blocks to allocate)
493 * @offsets: offsets (in the blocks) to store the pointers to next.
494 * @branch: place to store the chain in.
496 * This function allocates @num blocks, zeroes out all but the last one,
497 * links them into chain and (if we are synchronous) writes them to disk.
498 * In other words, it prepares a branch that can be spliced onto the
499 * inode. It stores the information about that chain in the branch[], in
500 * the same format as ext3_get_branch() would do. We are calling it after
501 * we had read the existing part of chain and partial points to the last
502 * triple of that (one with zero ->key). Upon the exit we have the same
503 * picture as after the successful ext3_get_block(), excpet that in one
504 * place chain is disconnected - *branch->p is still zero (we did not
505 * set the last link), but branch->key contains the number that should
506 * be placed into *branch->p to fill that gap.
508 * If allocation fails we free all blocks we've allocated (and forget
509 * their buffer_heads) and return the error value the from failed
510 * ext3_alloc_block() (normally -ENOSPC). Otherwise we set the chain
511 * as described above and return 0.
514 static int ext3_alloc_branch(handle_t *handle, struct inode *inode,
520 int blocksize = inode->i_sb->s_blocksize;
524 int parent = ext3_alloc_block(handle, inode, goal, &err);
526 branch[0].key = cpu_to_le32(parent);
528 for (n = 1; n < num; n++) {
529 struct buffer_head *bh;
530 /* Allocate the next block */
531 int nr = ext3_alloc_block(handle, inode, parent, &err);
534 branch[n].key = cpu_to_le32(nr);
538 * Get buffer_head for parent block, zero it out
539 * and set the pointer to new one, then send
542 bh = sb_getblk(inode->i_sb, parent);
545 BUFFER_TRACE(bh, "call get_create_access");
546 err = ext3_journal_get_create_access(handle, bh);
553 memset(bh->b_data, 0, blocksize);
554 branch[n].p = (__le32*) bh->b_data + offsets[n];
555 *branch[n].p = branch[n].key;
556 BUFFER_TRACE(bh, "marking uptodate");
557 set_buffer_uptodate(bh);
560 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
561 err = ext3_journal_dirty_metadata(handle, bh);
571 /* Allocation failed, free what we already allocated */
572 for (i = 1; i < keys; i++) {
573 BUFFER_TRACE(branch[i].bh, "call journal_forget");
574 ext3_journal_forget(handle, branch[i].bh);
576 for (i = 0; i < keys; i++)
577 ext3_free_blocks(handle, inode, le32_to_cpu(branch[i].key), 1);
582 * ext3_splice_branch - splice the allocated branch onto inode.
584 * @block: (logical) number of block we are adding
585 * @chain: chain of indirect blocks (with a missing link - see
587 * @where: location of missing link
588 * @num: number of blocks we are adding
590 * This function verifies that chain (up to the missing link) had not
591 * changed, fills the missing link and does all housekeeping needed in
592 * inode (->i_blocks, etc.). In case of success we end up with the full
593 * chain to new block and return 0. Otherwise (== chain had been changed)
594 * we free the new blocks (forgetting their buffer_heads, indeed) and
598 static int ext3_splice_branch(handle_t *handle, struct inode *inode, long block,
599 Indirect chain[4], Indirect *where, int num)
603 struct ext3_inode_info *ei = EXT3_I(inode);
606 * If we're splicing into a [td]indirect block (as opposed to the
607 * inode) then we need to get write access to the [td]indirect block
611 BUFFER_TRACE(where->bh, "get_write_access");
612 err = ext3_journal_get_write_access(handle, where->bh);
616 /* Verify that place we are splicing to is still there and vacant */
618 /* Writer: pointers, ->i_next_alloc* */
619 if (!verify_chain(chain, where-1) || *where->p)
625 *where->p = where->key;
626 ei->i_next_alloc_block = block;
627 ei->i_next_alloc_goal = le32_to_cpu(where[num-1].key);
630 /* We are done with atomic stuff, now do the rest of housekeeping */
632 inode->i_ctime = CURRENT_TIME;
633 ext3_mark_inode_dirty(handle, inode);
635 /* had we spliced it onto indirect block? */
638 * akpm: If we spliced it onto an indirect block, we haven't
639 * altered the inode. Note however that if it is being spliced
640 * onto an indirect block at the very end of the file (the
641 * file is growing) then we *will* alter the inode to reflect
642 * the new i_size. But that is not done here - it is done in
643 * generic_commit_write->__mark_inode_dirty->ext3_dirty_inode.
645 jbd_debug(5, "splicing indirect only\n");
646 BUFFER_TRACE(where->bh, "call ext3_journal_dirty_metadata");
647 err = ext3_journal_dirty_metadata(handle, where->bh);
652 * OK, we spliced it into the inode itself on a direct block.
653 * Inode was dirtied above.
655 jbd_debug(5, "splicing direct\n");
661 * AKPM: if where[i].bh isn't part of the current updating
662 * transaction then we explode nastily. Test this code path.
664 jbd_debug(1, "the chain changed: try again\n");
668 for (i = 1; i < num; i++) {
669 BUFFER_TRACE(where[i].bh, "call journal_forget");
670 ext3_journal_forget(handle, where[i].bh);
672 /* For the normal collision cleanup case, we free up the blocks.
673 * On genuine filesystem errors we don't even think about doing
676 for (i = 0; i < num; i++)
677 ext3_free_blocks(handle, inode,
678 le32_to_cpu(where[i].key), 1);
683 * Allocation strategy is simple: if we have to allocate something, we will
684 * have to go the whole way to leaf. So let's do it before attaching anything
685 * to tree, set linkage between the newborn blocks, write them if sync is
686 * required, recheck the path, free and repeat if check fails, otherwise
687 * set the last missing link (that will protect us from any truncate-generated
688 * removals - all blocks on the path are immune now) and possibly force the
689 * write on the parent block.
690 * That has a nice additional property: no special recovery from the failed
691 * allocations is needed - we simply release blocks and do not touch anything
692 * reachable from inode.
694 * akpm: `handle' can be NULL if create == 0.
696 * The BKL may not be held on entry here. Be sure to take it early.
700 ext3_get_block_handle(handle_t *handle, struct inode *inode, sector_t iblock,
701 struct buffer_head *bh_result, int create, int extend_disksize)
710 int depth = ext3_block_to_path(inode, iblock, offsets, &boundary);
711 struct ext3_inode_info *ei = EXT3_I(inode);
713 J_ASSERT(handle != NULL || create == 0);
719 partial = ext3_get_branch(inode, depth, offsets, chain, &err);
721 /* Simplest case - block found, no allocation needed */
723 clear_buffer_new(bh_result);
725 map_bh(bh_result, inode->i_sb, le32_to_cpu(chain[depth-1].key));
727 set_buffer_boundary(bh_result);
728 /* Clean up and exit */
729 partial = chain+depth-1; /* the whole chain */
733 /* Next simple case - plain lookup or failed read of indirect block */
734 if (!create || err == -EIO) {
736 while (partial > chain) {
737 BUFFER_TRACE(partial->bh, "call brelse");
741 BUFFER_TRACE(bh_result, "returned");
747 * Indirect block might be removed by truncate while we were
748 * reading it. Handling of that case (forget what we've got and
749 * reread) is taken out of the main path.
755 down(&ei->truncate_sem);
756 if (ext3_find_goal(inode, iblock, chain, partial, &goal) < 0) {
757 up(&ei->truncate_sem);
761 left = (chain + depth) - partial;
764 * Block out ext3_truncate while we alter the tree
766 err = ext3_alloc_branch(handle, inode, left, goal,
767 offsets+(partial-chain), partial);
769 /* The ext3_splice_branch call will free and forget any buffers
770 * on the new chain if there is a failure, but that risks using
771 * up transaction credits, especially for bitmaps where the
772 * credits cannot be returned. Can we handle this somehow? We
773 * may need to return -EAGAIN upwards in the worst case. --sct */
775 err = ext3_splice_branch(handle, inode, iblock, chain,
777 /* i_disksize growing is protected by truncate_sem
778 * don't forget to protect it if you're about to implement
779 * concurrent ext3_get_block() -bzzz */
780 if (!err && extend_disksize && inode->i_size > ei->i_disksize)
781 ei->i_disksize = inode->i_size;
782 up(&ei->truncate_sem);
788 set_buffer_new(bh_result);
792 while (partial > chain) {
793 jbd_debug(1, "buffer chain changed, retrying\n");
794 BUFFER_TRACE(partial->bh, "brelsing");
801 static int ext3_get_block(struct inode *inode, sector_t iblock,
802 struct buffer_head *bh_result, int create)
804 handle_t *handle = NULL;
808 handle = ext3_journal_current_handle();
809 J_ASSERT(handle != 0);
811 ret = ext3_get_block_handle(handle, inode, iblock,
812 bh_result, create, 1);
816 #define DIO_CREDITS (EXT3_RESERVE_TRANS_BLOCKS + 32)
819 ext3_direct_io_get_blocks(struct inode *inode, sector_t iblock,
820 unsigned long max_blocks, struct buffer_head *bh_result,
823 handle_t *handle = journal_current_handle();
827 goto get_block; /* A read */
829 if (handle->h_transaction->t_state == T_LOCKED) {
831 * Huge direct-io writes can hold off commits for long
832 * periods of time. Let this commit run.
834 ext3_journal_stop(handle);
835 handle = ext3_journal_start(inode, DIO_CREDITS);
837 ret = PTR_ERR(handle);
841 if (handle->h_buffer_credits <= EXT3_RESERVE_TRANS_BLOCKS) {
843 * Getting low on buffer credits...
845 ret = ext3_journal_extend(handle, DIO_CREDITS);
848 * Couldn't extend the transaction. Start a new one.
850 ret = ext3_journal_restart(handle, DIO_CREDITS);
856 ret = ext3_get_block_handle(handle, inode, iblock,
857 bh_result, create, 0);
858 bh_result->b_size = (1 << inode->i_blkbits);
863 * `handle' can be NULL if create is zero
865 struct buffer_head *ext3_getblk(handle_t *handle, struct inode * inode,
866 long block, int create, int * errp)
868 struct buffer_head dummy;
871 J_ASSERT(handle != NULL || create == 0);
874 dummy.b_blocknr = -1000;
875 buffer_trace_init(&dummy.b_history);
876 *errp = ext3_get_block_handle(handle, inode, block, &dummy, create, 1);
877 if (!*errp && buffer_mapped(&dummy)) {
878 struct buffer_head *bh;
879 bh = sb_getblk(inode->i_sb, dummy.b_blocknr);
880 if (buffer_new(&dummy)) {
881 J_ASSERT(create != 0);
882 J_ASSERT(handle != 0);
884 /* Now that we do not always journal data, we
885 should keep in mind whether this should
886 always journal the new buffer as metadata.
887 For now, regular file writes use
888 ext3_get_block instead, so it's not a
891 BUFFER_TRACE(bh, "call get_create_access");
892 fatal = ext3_journal_get_create_access(handle, bh);
893 if (!fatal && !buffer_uptodate(bh)) {
894 memset(bh->b_data, 0, inode->i_sb->s_blocksize);
895 set_buffer_uptodate(bh);
898 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
899 err = ext3_journal_dirty_metadata(handle, bh);
903 BUFFER_TRACE(bh, "not a new buffer");
915 struct buffer_head *ext3_bread(handle_t *handle, struct inode * inode,
916 int block, int create, int *err)
918 struct buffer_head * bh;
921 prev_blocks = inode->i_blocks;
923 bh = ext3_getblk (handle, inode, block, create, err);
926 if (buffer_uptodate(bh))
928 ll_rw_block (READ, 1, &bh);
930 if (buffer_uptodate(bh))
937 static int walk_page_buffers( handle_t *handle,
938 struct buffer_head *head,
942 int (*fn)( handle_t *handle,
943 struct buffer_head *bh))
945 struct buffer_head *bh;
946 unsigned block_start, block_end;
947 unsigned blocksize = head->b_size;
949 struct buffer_head *next;
951 for ( bh = head, block_start = 0;
952 ret == 0 && (bh != head || !block_start);
953 block_start = block_end, bh = next)
955 next = bh->b_this_page;
956 block_end = block_start + blocksize;
957 if (block_end <= from || block_start >= to) {
958 if (partial && !buffer_uptodate(bh))
962 err = (*fn)(handle, bh);
970 * To preserve ordering, it is essential that the hole instantiation and
971 * the data write be encapsulated in a single transaction. We cannot
972 * close off a transaction and start a new one between the ext3_get_block()
973 * and the commit_write(). So doing the journal_start at the start of
974 * prepare_write() is the right place.
976 * Also, this function can nest inside ext3_writepage() ->
977 * block_write_full_page(). In that case, we *know* that ext3_writepage()
978 * has generated enough buffer credits to do the whole page. So we won't
979 * block on the journal in that case, which is good, because the caller may
982 * By accident, ext3 can be reentered when a transaction is open via
983 * quota file writes. If we were to commit the transaction while thus
984 * reentered, there can be a deadlock - we would be holding a quota
985 * lock, and the commit would never complete if another thread had a
986 * transaction open and was blocking on the quota lock - a ranking
989 * So what we do is to rely on the fact that journal_stop/journal_start
990 * will _not_ run commit under these circumstances because handle->h_ref
991 * is elevated. We'll still have enough credits for the tiny quotafile
995 static int do_journal_get_write_access(handle_t *handle,
996 struct buffer_head *bh)
998 if (!buffer_mapped(bh) || buffer_freed(bh))
1000 return ext3_journal_get_write_access(handle, bh);
1003 static int ext3_prepare_write(struct file *file, struct page *page,
1004 unsigned from, unsigned to)
1006 struct inode *inode = page->mapping->host;
1007 int ret, needed_blocks = ext3_writepage_trans_blocks(inode);
1012 handle = ext3_journal_start(inode, needed_blocks);
1013 if (IS_ERR(handle)) {
1014 ret = PTR_ERR(handle);
1017 ret = block_prepare_write(page, from, to, ext3_get_block);
1019 goto prepare_write_failed;
1021 if (ext3_should_journal_data(inode)) {
1022 ret = walk_page_buffers(handle, page_buffers(page),
1023 from, to, NULL, do_journal_get_write_access);
1025 prepare_write_failed:
1027 ext3_journal_stop(handle);
1028 if (ret == -ENOSPC && ext3_should_retry_alloc(inode->i_sb, &retries))
1035 ext3_journal_dirty_data(handle_t *handle, struct buffer_head *bh)
1037 int err = journal_dirty_data(handle, bh);
1039 ext3_journal_abort_handle(__FUNCTION__, __FUNCTION__,
1044 /* For commit_write() in data=journal mode */
1045 static int commit_write_fn(handle_t *handle, struct buffer_head *bh)
1047 if (!buffer_mapped(bh) || buffer_freed(bh))
1049 set_buffer_uptodate(bh);
1050 return ext3_journal_dirty_metadata(handle, bh);
1054 * We need to pick up the new inode size which generic_commit_write gave us
1055 * `file' can be NULL - eg, when called from page_symlink().
1057 * ext3 never places buffers on inode->i_mapping->private_list. metadata
1058 * buffers are managed internally.
1061 static int ext3_ordered_commit_write(struct file *file, struct page *page,
1062 unsigned from, unsigned to)
1064 handle_t *handle = ext3_journal_current_handle();
1065 struct inode *inode = page->mapping->host;
1068 ret = walk_page_buffers(handle, page_buffers(page),
1069 from, to, NULL, ext3_journal_dirty_data);
1073 * generic_commit_write() will run mark_inode_dirty() if i_size
1074 * changes. So let's piggyback the i_disksize mark_inode_dirty
1079 new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1080 if (new_i_size > EXT3_I(inode)->i_disksize)
1081 EXT3_I(inode)->i_disksize = new_i_size;
1082 ret = generic_commit_write(file, page, from, to);
1084 ret2 = ext3_journal_stop(handle);
1090 static int ext3_writeback_commit_write(struct file *file, struct page *page,
1091 unsigned from, unsigned to)
1093 handle_t *handle = ext3_journal_current_handle();
1094 struct inode *inode = page->mapping->host;
1098 new_i_size = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1099 if (new_i_size > EXT3_I(inode)->i_disksize)
1100 EXT3_I(inode)->i_disksize = new_i_size;
1101 ret = generic_commit_write(file, page, from, to);
1102 ret2 = ext3_journal_stop(handle);
1108 static int ext3_journalled_commit_write(struct file *file,
1109 struct page *page, unsigned from, unsigned to)
1111 handle_t *handle = ext3_journal_current_handle();
1112 struct inode *inode = page->mapping->host;
1118 * Here we duplicate the generic_commit_write() functionality
1120 pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
1122 ret = walk_page_buffers(handle, page_buffers(page), from,
1123 to, &partial, commit_write_fn);
1125 SetPageUptodate(page);
1126 if (pos > inode->i_size)
1127 i_size_write(inode, pos);
1128 EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
1129 if (inode->i_size > EXT3_I(inode)->i_disksize) {
1130 EXT3_I(inode)->i_disksize = inode->i_size;
1131 ret2 = ext3_mark_inode_dirty(handle, inode);
1135 ret2 = ext3_journal_stop(handle);
1142 * bmap() is special. It gets used by applications such as lilo and by
1143 * the swapper to find the on-disk block of a specific piece of data.
1145 * Naturally, this is dangerous if the block concerned is still in the
1146 * journal. If somebody makes a swapfile on an ext3 data-journaling
1147 * filesystem and enables swap, then they may get a nasty shock when the
1148 * data getting swapped to that swapfile suddenly gets overwritten by
1149 * the original zero's written out previously to the journal and
1150 * awaiting writeback in the kernel's buffer cache.
1152 * So, if we see any bmap calls here on a modified, data-journaled file,
1153 * take extra steps to flush any blocks which might be in the cache.
1155 static sector_t ext3_bmap(struct address_space *mapping, sector_t block)
1157 struct inode *inode = mapping->host;
1161 if (EXT3_I(inode)->i_state & EXT3_STATE_JDATA) {
1163 * This is a REALLY heavyweight approach, but the use of
1164 * bmap on dirty files is expected to be extremely rare:
1165 * only if we run lilo or swapon on a freshly made file
1166 * do we expect this to happen.
1168 * (bmap requires CAP_SYS_RAWIO so this does not
1169 * represent an unprivileged user DOS attack --- we'd be
1170 * in trouble if mortal users could trigger this path at
1173 * NB. EXT3_STATE_JDATA is not set on files other than
1174 * regular files. If somebody wants to bmap a directory
1175 * or symlink and gets confused because the buffer
1176 * hasn't yet been flushed to disk, they deserve
1177 * everything they get.
1180 EXT3_I(inode)->i_state &= ~EXT3_STATE_JDATA;
1181 journal = EXT3_JOURNAL(inode);
1182 journal_lock_updates(journal);
1183 err = journal_flush(journal);
1184 journal_unlock_updates(journal);
1190 return generic_block_bmap(mapping,block,ext3_get_block);
1193 static int bget_one(handle_t *handle, struct buffer_head *bh)
1199 static int bput_one(handle_t *handle, struct buffer_head *bh)
1205 static int journal_dirty_data_fn(handle_t *handle, struct buffer_head *bh)
1207 if (buffer_mapped(bh))
1208 return ext3_journal_dirty_data(handle, bh);
1213 * Note that we always start a transaction even if we're not journalling
1214 * data. This is to preserve ordering: any hole instantiation within
1215 * __block_write_full_page -> ext3_get_block() should be journalled
1216 * along with the data so we don't crash and then get metadata which
1217 * refers to old data.
1219 * In all journalling modes block_write_full_page() will start the I/O.
1223 * ext3_writepage() -> kmalloc() -> __alloc_pages() -> page_launder() ->
1228 * ext3_file_write() -> generic_file_write() -> __alloc_pages() -> ...
1230 * Same applies to ext3_get_block(). We will deadlock on various things like
1231 * lock_journal and i_truncate_sem.
1233 * Setting PF_MEMALLOC here doesn't work - too many internal memory
1236 * 16May01: If we're reentered then journal_current_handle() will be
1237 * non-zero. We simply *return*.
1239 * 1 July 2001: @@@ FIXME:
1240 * In journalled data mode, a data buffer may be metadata against the
1241 * current transaction. But the same file is part of a shared mapping
1242 * and someone does a writepage() on it.
1244 * We will move the buffer onto the async_data list, but *after* it has
1245 * been dirtied. So there's a small window where we have dirty data on
1248 * Note that this only applies to the last partial page in the file. The
1249 * bit which block_write_full_page() uses prepare/commit for. (That's
1250 * broken code anyway: it's wrong for msync()).
1252 * It's a rare case: affects the final partial page, for journalled data
1253 * where the file is subject to bith write() and writepage() in the same
1254 * transction. To fix it we'll need a custom block_write_full_page().
1255 * We'll probably need that anyway for journalling writepage() output.
1257 * We don't honour synchronous mounts for writepage(). That would be
1258 * disastrous. Any write() or metadata operation will sync the fs for
1261 * AKPM2: if all the page's buffers are mapped to disk and !data=journal,
1262 * we don't need to open a transaction here.
1264 static int ext3_ordered_writepage(struct page *page,
1265 struct writeback_control *wbc)
1267 struct inode *inode = page->mapping->host;
1268 struct buffer_head *page_bufs;
1269 handle_t *handle = NULL;
1273 J_ASSERT(PageLocked(page));
1276 * We give up here if we're reentered, because it might be for a
1277 * different filesystem.
1279 if (ext3_journal_current_handle())
1282 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1284 if (IS_ERR(handle)) {
1285 ret = PTR_ERR(handle);
1289 if (!page_has_buffers(page)) {
1290 create_empty_buffers(page, inode->i_sb->s_blocksize,
1291 (1 << BH_Dirty)|(1 << BH_Uptodate));
1293 page_bufs = page_buffers(page);
1294 walk_page_buffers(handle, page_bufs, 0,
1295 PAGE_CACHE_SIZE, NULL, bget_one);
1297 ret = block_write_full_page(page, ext3_get_block, wbc);
1300 * The page can become unlocked at any point now, and
1301 * truncate can then come in and change things. So we
1302 * can't touch *page from now on. But *page_bufs is
1303 * safe due to elevated refcount.
1307 * And attach them to the current transaction. But only if
1308 * block_write_full_page() succeeded. Otherwise they are unmapped,
1309 * and generally junk.
1312 err = walk_page_buffers(handle, page_bufs, 0, PAGE_CACHE_SIZE,
1313 NULL, journal_dirty_data_fn);
1317 walk_page_buffers(handle, page_bufs, 0,
1318 PAGE_CACHE_SIZE, NULL, bput_one);
1319 err = ext3_journal_stop(handle);
1325 redirty_page_for_writepage(wbc, page);
1330 static int ext3_writeback_writepage(struct page *page,
1331 struct writeback_control *wbc)
1333 struct inode *inode = page->mapping->host;
1334 handle_t *handle = NULL;
1338 if (ext3_journal_current_handle())
1341 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1342 if (IS_ERR(handle)) {
1343 ret = PTR_ERR(handle);
1347 ret = block_write_full_page(page, ext3_get_block, wbc);
1348 err = ext3_journal_stop(handle);
1354 redirty_page_for_writepage(wbc, page);
1359 static int ext3_journalled_writepage(struct page *page,
1360 struct writeback_control *wbc)
1362 struct inode *inode = page->mapping->host;
1363 handle_t *handle = NULL;
1367 if (ext3_journal_current_handle())
1370 handle = ext3_journal_start(inode, ext3_writepage_trans_blocks(inode));
1371 if (IS_ERR(handle)) {
1372 ret = PTR_ERR(handle);
1376 if (!page_has_buffers(page) || PageChecked(page)) {
1378 * It's mmapped pagecache. Add buffers and journal it. There
1379 * doesn't seem much point in redirtying the page here.
1381 ClearPageChecked(page);
1382 ret = block_prepare_write(page, 0, PAGE_CACHE_SIZE,
1386 ret = walk_page_buffers(handle, page_buffers(page), 0,
1387 PAGE_CACHE_SIZE, NULL, do_journal_get_write_access);
1389 err = walk_page_buffers(handle, page_buffers(page), 0,
1390 PAGE_CACHE_SIZE, NULL, commit_write_fn);
1393 EXT3_I(inode)->i_state |= EXT3_STATE_JDATA;
1397 * It may be a page full of checkpoint-mode buffers. We don't
1398 * really know unless we go poke around in the buffer_heads.
1399 * But block_write_full_page will do the right thing.
1401 ret = block_write_full_page(page, ext3_get_block, wbc);
1403 err = ext3_journal_stop(handle);
1410 redirty_page_for_writepage(wbc, page);
1416 static int ext3_readpage(struct file *file, struct page *page)
1418 return mpage_readpage(page, ext3_get_block);
1422 ext3_readpages(struct file *file, struct address_space *mapping,
1423 struct list_head *pages, unsigned nr_pages)
1425 return mpage_readpages(mapping, pages, nr_pages, ext3_get_block);
1428 static int ext3_invalidatepage(struct page *page, unsigned long offset)
1430 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1433 * If it's a full truncate we just forget about the pending dirtying
1436 ClearPageChecked(page);
1438 return journal_invalidatepage(journal, page, offset);
1441 static int ext3_releasepage(struct page *page, int wait)
1443 journal_t *journal = EXT3_JOURNAL(page->mapping->host);
1445 WARN_ON(PageChecked(page));
1446 return journal_try_to_free_buffers(journal, page, wait);
1450 * If the O_DIRECT write will extend the file then add this inode to the
1451 * orphan list. So recovery will truncate it back to the original size
1452 * if the machine crashes during the write.
1454 * If the O_DIRECT write is intantiating holes inside i_size and the machine
1455 * crashes then stale disk data _may_ be exposed inside the file.
1457 static ssize_t ext3_direct_IO(int rw, struct kiocb *iocb,
1458 const struct iovec *iov, loff_t offset,
1459 unsigned long nr_segs)
1461 struct file *file = iocb->ki_filp;
1462 struct inode *inode = file->f_mapping->host;
1463 struct ext3_inode_info *ei = EXT3_I(inode);
1464 handle_t *handle = NULL;
1467 size_t count = iov_length(iov, nr_segs);
1470 loff_t final_size = offset + count;
1472 handle = ext3_journal_start(inode, DIO_CREDITS);
1473 if (IS_ERR(handle)) {
1474 ret = PTR_ERR(handle);
1477 if (final_size > inode->i_size) {
1478 ret = ext3_orphan_add(handle, inode);
1482 ei->i_disksize = inode->i_size;
1486 ret = blockdev_direct_IO(rw, iocb, inode, inode->i_sb->s_bdev, iov,
1488 ext3_direct_io_get_blocks, NULL);
1491 * Reacquire the handle: ext3_direct_io_get_block() can restart the
1494 handle = journal_current_handle();
1501 ext3_orphan_del(handle, inode);
1502 if (orphan && ret > 0) {
1503 loff_t end = offset + ret;
1504 if (end > inode->i_size) {
1505 ei->i_disksize = end;
1506 i_size_write(inode, end);
1507 err = ext3_mark_inode_dirty(handle, inode);
1512 err = ext3_journal_stop(handle);
1521 * Pages can be marked dirty completely asynchronously from ext3's journalling
1522 * activity. By filemap_sync_pte(), try_to_unmap_one(), etc. We cannot do
1523 * much here because ->set_page_dirty is called under VFS locks. The page is
1524 * not necessarily locked.
1526 * We cannot just dirty the page and leave attached buffers clean, because the
1527 * buffers' dirty state is "definitive". We cannot just set the buffers dirty
1528 * or jbddirty because all the journalling code will explode.
1530 * So what we do is to mark the page "pending dirty" and next time writepage
1531 * is called, propagate that into the buffers appropriately.
1533 static int ext3_journalled_set_page_dirty(struct page *page)
1535 SetPageChecked(page);
1536 return __set_page_dirty_nobuffers(page);
1539 static struct address_space_operations ext3_ordered_aops = {
1540 .readpage = ext3_readpage,
1541 .readpages = ext3_readpages,
1542 .writepage = ext3_ordered_writepage,
1543 .sync_page = block_sync_page,
1544 .prepare_write = ext3_prepare_write,
1545 .commit_write = ext3_ordered_commit_write,
1547 .invalidatepage = ext3_invalidatepage,
1548 .releasepage = ext3_releasepage,
1549 .direct_IO = ext3_direct_IO,
1552 static struct address_space_operations ext3_writeback_aops = {
1553 .readpage = ext3_readpage,
1554 .readpages = ext3_readpages,
1555 .writepage = ext3_writeback_writepage,
1556 .sync_page = block_sync_page,
1557 .prepare_write = ext3_prepare_write,
1558 .commit_write = ext3_writeback_commit_write,
1560 .invalidatepage = ext3_invalidatepage,
1561 .releasepage = ext3_releasepage,
1562 .direct_IO = ext3_direct_IO,
1565 static struct address_space_operations ext3_journalled_aops = {
1566 .readpage = ext3_readpage,
1567 .readpages = ext3_readpages,
1568 .writepage = ext3_journalled_writepage,
1569 .sync_page = block_sync_page,
1570 .prepare_write = ext3_prepare_write,
1571 .commit_write = ext3_journalled_commit_write,
1572 .set_page_dirty = ext3_journalled_set_page_dirty,
1574 .invalidatepage = ext3_invalidatepage,
1575 .releasepage = ext3_releasepage,
1578 void ext3_set_aops(struct inode *inode)
1580 if (ext3_should_order_data(inode))
1581 inode->i_mapping->a_ops = &ext3_ordered_aops;
1582 else if (ext3_should_writeback_data(inode))
1583 inode->i_mapping->a_ops = &ext3_writeback_aops;
1585 inode->i_mapping->a_ops = &ext3_journalled_aops;
1589 * ext3_block_truncate_page() zeroes out a mapping from file offset `from'
1590 * up to the end of the block which corresponds to `from'.
1591 * This required during truncate. We need to physically zero the tail end
1592 * of that block so it doesn't yield old data if the file is later grown.
1594 static int ext3_block_truncate_page(handle_t *handle, struct page *page,
1595 struct address_space *mapping, loff_t from)
1597 unsigned long index = from >> PAGE_CACHE_SHIFT;
1598 unsigned offset = from & (PAGE_CACHE_SIZE-1);
1599 unsigned blocksize, iblock, length, pos;
1600 struct inode *inode = mapping->host;
1601 struct buffer_head *bh;
1605 blocksize = inode->i_sb->s_blocksize;
1606 length = blocksize - (offset & (blocksize - 1));
1607 iblock = index << (PAGE_CACHE_SHIFT - inode->i_sb->s_blocksize_bits);
1609 if (!page_has_buffers(page))
1610 create_empty_buffers(page, blocksize, 0);
1612 /* Find the buffer that contains "offset" */
1613 bh = page_buffers(page);
1615 while (offset >= pos) {
1616 bh = bh->b_this_page;
1622 if (buffer_freed(bh)) {
1623 BUFFER_TRACE(bh, "freed: skip");
1627 if (!buffer_mapped(bh)) {
1628 BUFFER_TRACE(bh, "unmapped");
1629 ext3_get_block(inode, iblock, bh, 0);
1630 /* unmapped? It's a hole - nothing to do */
1631 if (!buffer_mapped(bh)) {
1632 BUFFER_TRACE(bh, "still unmapped");
1637 /* Ok, it's mapped. Make sure it's up-to-date */
1638 if (PageUptodate(page))
1639 set_buffer_uptodate(bh);
1641 if (!buffer_uptodate(bh)) {
1643 ll_rw_block(READ, 1, &bh);
1645 /* Uhhuh. Read error. Complain and punt. */
1646 if (!buffer_uptodate(bh))
1650 if (ext3_should_journal_data(inode)) {
1651 BUFFER_TRACE(bh, "get write access");
1652 err = ext3_journal_get_write_access(handle, bh);
1657 kaddr = kmap_atomic(page, KM_USER0);
1658 memset(kaddr + offset, 0, length);
1659 flush_dcache_page(page);
1660 kunmap_atomic(kaddr, KM_USER0);
1662 BUFFER_TRACE(bh, "zeroed end of block");
1665 if (ext3_should_journal_data(inode)) {
1666 err = ext3_journal_dirty_metadata(handle, bh);
1668 if (ext3_should_order_data(inode))
1669 err = ext3_journal_dirty_data(handle, bh);
1670 mark_buffer_dirty(bh);
1675 page_cache_release(page);
1680 * Probably it should be a library function... search for first non-zero word
1681 * or memcmp with zero_page, whatever is better for particular architecture.
1684 static inline int all_zeroes(__le32 *p, __le32 *q)
1693 * ext3_find_shared - find the indirect blocks for partial truncation.
1694 * @inode: inode in question
1695 * @depth: depth of the affected branch
1696 * @offsets: offsets of pointers in that branch (see ext3_block_to_path)
1697 * @chain: place to store the pointers to partial indirect blocks
1698 * @top: place to the (detached) top of branch
1700 * This is a helper function used by ext3_truncate().
1702 * When we do truncate() we may have to clean the ends of several
1703 * indirect blocks but leave the blocks themselves alive. Block is
1704 * partially truncated if some data below the new i_size is refered
1705 * from it (and it is on the path to the first completely truncated
1706 * data block, indeed). We have to free the top of that path along
1707 * with everything to the right of the path. Since no allocation
1708 * past the truncation point is possible until ext3_truncate()
1709 * finishes, we may safely do the latter, but top of branch may
1710 * require special attention - pageout below the truncation point
1711 * might try to populate it.
1713 * We atomically detach the top of branch from the tree, store the
1714 * block number of its root in *@top, pointers to buffer_heads of
1715 * partially truncated blocks - in @chain[].bh and pointers to
1716 * their last elements that should not be removed - in
1717 * @chain[].p. Return value is the pointer to last filled element
1720 * The work left to caller to do the actual freeing of subtrees:
1721 * a) free the subtree starting from *@top
1722 * b) free the subtrees whose roots are stored in
1723 * (@chain[i].p+1 .. end of @chain[i].bh->b_data)
1724 * c) free the subtrees growing from the inode past the @chain[0].
1725 * (no partially truncated stuff there). */
1727 static Indirect *ext3_find_shared(struct inode *inode,
1733 Indirect *partial, *p;
1737 /* Make k index the deepest non-null offest + 1 */
1738 for (k = depth; k > 1 && !offsets[k-1]; k--)
1740 partial = ext3_get_branch(inode, k, offsets, chain, &err);
1741 /* Writer: pointers */
1743 partial = chain + k-1;
1745 * If the branch acquired continuation since we've looked at it -
1746 * fine, it should all survive and (new) top doesn't belong to us.
1748 if (!partial->key && *partial->p)
1751 for (p=partial; p>chain && all_zeroes((__le32*)p->bh->b_data,p->p); p--)
1754 * OK, we've found the last block that must survive. The rest of our
1755 * branch should be detached before unlocking. However, if that rest
1756 * of branch is all ours and does not grow immediately from the inode
1757 * it's easier to cheat and just decrement partial->p.
1759 if (p == chain + k - 1 && p > chain) {
1763 /* Nope, don't do this in ext3. Must leave the tree intact */
1772 brelse(partial->bh);
1780 * Zero a number of block pointers in either an inode or an indirect block.
1781 * If we restart the transaction we must again get write access to the
1782 * indirect block for further modification.
1784 * We release `count' blocks on disk, but (last - first) may be greater
1785 * than `count' because there can be holes in there.
1788 ext3_clear_blocks(handle_t *handle, struct inode *inode, struct buffer_head *bh,
1789 unsigned long block_to_free, unsigned long count,
1790 __le32 *first, __le32 *last)
1793 if (try_to_extend_transaction(handle, inode)) {
1795 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
1796 ext3_journal_dirty_metadata(handle, bh);
1798 ext3_mark_inode_dirty(handle, inode);
1799 ext3_journal_test_restart(handle, inode);
1801 BUFFER_TRACE(bh, "retaking write access");
1802 ext3_journal_get_write_access(handle, bh);
1807 * Any buffers which are on the journal will be in memory. We find
1808 * them on the hash table so journal_revoke() will run journal_forget()
1809 * on them. We've already detached each block from the file, so
1810 * bforget() in journal_forget() should be safe.
1812 * AKPM: turn on bforget in journal_forget()!!!
1814 for (p = first; p < last; p++) {
1815 u32 nr = le32_to_cpu(*p);
1817 struct buffer_head *bh;
1820 bh = sb_find_get_block(inode->i_sb, nr);
1821 ext3_forget(handle, 0, inode, bh, nr);
1825 ext3_free_blocks(handle, inode, block_to_free, count);
1829 * ext3_free_data - free a list of data blocks
1830 * @handle: handle for this transaction
1831 * @inode: inode we are dealing with
1832 * @this_bh: indirect buffer_head which contains *@first and *@last
1833 * @first: array of block numbers
1834 * @last: points immediately past the end of array
1836 * We are freeing all blocks refered from that array (numbers are stored as
1837 * little-endian 32-bit) and updating @inode->i_blocks appropriately.
1839 * We accumulate contiguous runs of blocks to free. Conveniently, if these
1840 * blocks are contiguous then releasing them at one time will only affect one
1841 * or two bitmap blocks (+ group descriptor(s) and superblock) and we won't
1842 * actually use a lot of journal space.
1844 * @this_bh will be %NULL if @first and @last point into the inode's direct
1847 static void ext3_free_data(handle_t *handle, struct inode *inode,
1848 struct buffer_head *this_bh,
1849 __le32 *first, __le32 *last)
1851 unsigned long block_to_free = 0; /* Starting block # of a run */
1852 unsigned long count = 0; /* Number of blocks in the run */
1853 __le32 *block_to_free_p = NULL; /* Pointer into inode/ind
1856 unsigned long nr; /* Current block # */
1857 __le32 *p; /* Pointer into inode/ind
1858 for current block */
1861 if (this_bh) { /* For indirect block */
1862 BUFFER_TRACE(this_bh, "get_write_access");
1863 err = ext3_journal_get_write_access(handle, this_bh);
1864 /* Important: if we can't update the indirect pointers
1865 * to the blocks, we can't free them. */
1870 for (p = first; p < last; p++) {
1871 nr = le32_to_cpu(*p);
1873 /* accumulate blocks to free if they're contiguous */
1876 block_to_free_p = p;
1878 } else if (nr == block_to_free + count) {
1881 ext3_clear_blocks(handle, inode, this_bh,
1883 count, block_to_free_p, p);
1885 block_to_free_p = p;
1892 ext3_clear_blocks(handle, inode, this_bh, block_to_free,
1893 count, block_to_free_p, p);
1896 BUFFER_TRACE(this_bh, "call ext3_journal_dirty_metadata");
1897 ext3_journal_dirty_metadata(handle, this_bh);
1902 * ext3_free_branches - free an array of branches
1903 * @handle: JBD handle for this transaction
1904 * @inode: inode we are dealing with
1905 * @parent_bh: the buffer_head which contains *@first and *@last
1906 * @first: array of block numbers
1907 * @last: pointer immediately past the end of array
1908 * @depth: depth of the branches to free
1910 * We are freeing all blocks refered from these branches (numbers are
1911 * stored as little-endian 32-bit) and updating @inode->i_blocks
1914 static void ext3_free_branches(handle_t *handle, struct inode *inode,
1915 struct buffer_head *parent_bh,
1916 __le32 *first, __le32 *last, int depth)
1921 if (is_handle_aborted(handle))
1925 struct buffer_head *bh;
1926 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
1928 while (--p >= first) {
1929 nr = le32_to_cpu(*p);
1931 continue; /* A hole */
1933 /* Go read the buffer for the next level down */
1934 bh = sb_bread(inode->i_sb, nr);
1937 * A read failure? Report error and clear slot
1941 ext3_error(inode->i_sb, "ext3_free_branches",
1942 "Read failure, inode=%ld, block=%ld",
1947 /* This zaps the entire block. Bottom up. */
1948 BUFFER_TRACE(bh, "free child branches");
1949 ext3_free_branches(handle, inode, bh,
1950 (__le32*)bh->b_data,
1951 (__le32*)bh->b_data + addr_per_block,
1955 * We've probably journalled the indirect block several
1956 * times during the truncate. But it's no longer
1957 * needed and we now drop it from the transaction via
1960 * That's easy if it's exclusively part of this
1961 * transaction. But if it's part of the committing
1962 * transaction then journal_forget() will simply
1963 * brelse() it. That means that if the underlying
1964 * block is reallocated in ext3_get_block(),
1965 * unmap_underlying_metadata() will find this block
1966 * and will try to get rid of it. damn, damn.
1968 * If this block has already been committed to the
1969 * journal, a revoke record will be written. And
1970 * revoke records must be emitted *before* clearing
1971 * this block's bit in the bitmaps.
1973 ext3_forget(handle, 1, inode, bh, bh->b_blocknr);
1976 * Everything below this this pointer has been
1977 * released. Now let this top-of-subtree go.
1979 * We want the freeing of this indirect block to be
1980 * atomic in the journal with the updating of the
1981 * bitmap block which owns it. So make some room in
1984 * We zero the parent pointer *after* freeing its
1985 * pointee in the bitmaps, so if extend_transaction()
1986 * for some reason fails to put the bitmap changes and
1987 * the release into the same transaction, recovery
1988 * will merely complain about releasing a free block,
1989 * rather than leaking blocks.
1991 if (is_handle_aborted(handle))
1993 if (try_to_extend_transaction(handle, inode)) {
1994 ext3_mark_inode_dirty(handle, inode);
1995 ext3_journal_test_restart(handle, inode);
1998 ext3_free_blocks(handle, inode, nr, 1);
2002 * The block which we have just freed is
2003 * pointed to by an indirect block: journal it
2005 BUFFER_TRACE(parent_bh, "get_write_access");
2006 if (!ext3_journal_get_write_access(handle,
2009 BUFFER_TRACE(parent_bh,
2010 "call ext3_journal_dirty_metadata");
2011 ext3_journal_dirty_metadata(handle,
2017 /* We have reached the bottom of the tree. */
2018 BUFFER_TRACE(parent_bh, "free data blocks");
2019 ext3_free_data(handle, inode, parent_bh, first, last);
2026 * We block out ext3_get_block() block instantiations across the entire
2027 * transaction, and VFS/VM ensures that ext3_truncate() cannot run
2028 * simultaneously on behalf of the same inode.
2030 * As we work through the truncate and commmit bits of it to the journal there
2031 * is one core, guiding principle: the file's tree must always be consistent on
2032 * disk. We must be able to restart the truncate after a crash.
2034 * The file's tree may be transiently inconsistent in memory (although it
2035 * probably isn't), but whenever we close off and commit a journal transaction,
2036 * the contents of (the filesystem + the journal) must be consistent and
2037 * restartable. It's pretty simple, really: bottom up, right to left (although
2038 * left-to-right works OK too).
2040 * Note that at recovery time, journal replay occurs *before* the restart of
2041 * truncate against the orphan inode list.
2043 * The committed inode has the new, desired i_size (which is the same as
2044 * i_disksize in this case). After a crash, ext3_orphan_cleanup() will see
2045 * that this inode's truncate did not complete and it will again call
2046 * ext3_truncate() to have another go. So there will be instantiated blocks
2047 * to the right of the truncation point in a crashed ext3 filesystem. But
2048 * that's fine - as long as they are linked from the inode, the post-crash
2049 * ext3_truncate() run will find them and release them.
2052 void ext3_truncate_nocheck(struct inode * inode)
2055 struct ext3_inode_info *ei = EXT3_I(inode);
2056 __le32 *i_data = ei->i_data;
2057 int addr_per_block = EXT3_ADDR_PER_BLOCK(inode->i_sb);
2058 struct address_space *mapping = inode->i_mapping;
2065 unsigned blocksize = inode->i_sb->s_blocksize;
2068 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) ||
2069 S_ISLNK(inode->i_mode)))
2071 if (ext3_inode_is_fast_symlink(inode))
2074 ext3_discard_reservation(inode);
2077 * We have to lock the EOF page here, because lock_page() nests
2078 * outside journal_start().
2080 if ((inode->i_size & (blocksize - 1)) == 0) {
2081 /* Block boundary? Nothing to do */
2084 page = grab_cache_page(mapping,
2085 inode->i_size >> PAGE_CACHE_SHIFT);
2090 handle = start_transaction(inode);
2091 if (IS_ERR(handle)) {
2093 clear_highpage(page);
2094 flush_dcache_page(page);
2096 page_cache_release(page);
2098 return; /* AKPM: return what? */
2101 last_block = (inode->i_size + blocksize-1)
2102 >> EXT3_BLOCK_SIZE_BITS(inode->i_sb);
2105 ext3_block_truncate_page(handle, page, mapping, inode->i_size);
2107 n = ext3_block_to_path(inode, last_block, offsets, NULL);
2109 goto out_stop; /* error */
2112 * OK. This truncate is going to happen. We add the inode to the
2113 * orphan list, so that if this truncate spans multiple transactions,
2114 * and we crash, we will resume the truncate when the filesystem
2115 * recovers. It also marks the inode dirty, to catch the new size.
2117 * Implication: the file must always be in a sane, consistent
2118 * truncatable state while each transaction commits.
2120 if (ext3_orphan_add(handle, inode))
2124 * The orphan list entry will now protect us from any crash which
2125 * occurs before the truncate completes, so it is now safe to propagate
2126 * the new, shorter inode size (held for now in i_size) into the
2127 * on-disk inode. We do this via i_disksize, which is the value which
2128 * ext3 *really* writes onto the disk inode.
2130 ei->i_disksize = inode->i_size;
2133 * From here we block out all ext3_get_block() callers who want to
2134 * modify the block allocation tree.
2136 down(&ei->truncate_sem);
2138 if (n == 1) { /* direct blocks */
2139 ext3_free_data(handle, inode, NULL, i_data+offsets[0],
2140 i_data + EXT3_NDIR_BLOCKS);
2144 partial = ext3_find_shared(inode, n, offsets, chain, &nr);
2145 /* Kill the top of shared branch (not detached) */
2147 if (partial == chain) {
2148 /* Shared branch grows from the inode */
2149 ext3_free_branches(handle, inode, NULL,
2150 &nr, &nr+1, (chain+n-1) - partial);
2153 * We mark the inode dirty prior to restart,
2154 * and prior to stop. No need for it here.
2157 /* Shared branch grows from an indirect block */
2158 BUFFER_TRACE(partial->bh, "get_write_access");
2159 ext3_free_branches(handle, inode, partial->bh,
2161 partial->p+1, (chain+n-1) - partial);
2164 /* Clear the ends of indirect blocks on the shared branch */
2165 while (partial > chain) {
2166 ext3_free_branches(handle, inode, partial->bh, partial->p + 1,
2167 (__le32*)partial->bh->b_data+addr_per_block,
2168 (chain+n-1) - partial);
2169 BUFFER_TRACE(partial->bh, "call brelse");
2170 brelse (partial->bh);
2174 /* Kill the remaining (whole) subtrees */
2175 switch (offsets[0]) {
2177 nr = i_data[EXT3_IND_BLOCK];
2179 ext3_free_branches(handle, inode, NULL,
2181 i_data[EXT3_IND_BLOCK] = 0;
2183 case EXT3_IND_BLOCK:
2184 nr = i_data[EXT3_DIND_BLOCK];
2186 ext3_free_branches(handle, inode, NULL,
2188 i_data[EXT3_DIND_BLOCK] = 0;
2190 case EXT3_DIND_BLOCK:
2191 nr = i_data[EXT3_TIND_BLOCK];
2193 ext3_free_branches(handle, inode, NULL,
2195 i_data[EXT3_TIND_BLOCK] = 0;
2197 case EXT3_TIND_BLOCK:
2200 up(&ei->truncate_sem);
2201 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
2202 ext3_mark_inode_dirty(handle, inode);
2204 /* In a multi-transaction truncate, we only make the final
2205 * transaction synchronous */
2210 * If this was a simple ftruncate(), and the file will remain alive
2211 * then we need to clear up the orphan record which we created above.
2212 * However, if this was a real unlink then we were called by
2213 * ext3_delete_inode(), and we allow that function to clean up the
2214 * orphan info for us.
2217 ext3_orphan_del(handle, inode);
2219 ext3_journal_stop(handle);
2222 static unsigned long ext3_get_inode_block(struct super_block *sb,
2223 unsigned long ino, struct ext3_iloc *iloc)
2225 unsigned long desc, group_desc, block_group;
2226 unsigned long offset, block;
2227 struct buffer_head *bh;
2228 struct ext3_group_desc * gdp;
2231 if ((ino != EXT3_ROOT_INO &&
2232 ino != EXT3_JOURNAL_INO &&
2233 ino != EXT3_RESIZE_INO &&
2234 ino < EXT3_FIRST_INO(sb)) ||
2236 EXT3_SB(sb)->s_es->s_inodes_count)) {
2237 ext3_error (sb, "ext3_get_inode_block",
2238 "bad inode number: %lu", ino);
2241 block_group = (ino - 1) / EXT3_INODES_PER_GROUP(sb);
2242 if (block_group >= EXT3_SB(sb)->s_groups_count) {
2243 ext3_error (sb, "ext3_get_inode_block",
2244 "group >= groups count");
2248 group_desc = block_group >> EXT3_DESC_PER_BLOCK_BITS(sb);
2249 desc = block_group & (EXT3_DESC_PER_BLOCK(sb) - 1);
2250 bh = EXT3_SB(sb)->s_group_desc[group_desc];
2252 ext3_error (sb, "ext3_get_inode_block",
2253 "Descriptor not loaded");
2257 gdp = (struct ext3_group_desc *) bh->b_data;
2259 * Figure out the offset within the block group inode table
2261 offset = ((ino - 1) % EXT3_INODES_PER_GROUP(sb)) *
2262 EXT3_INODE_SIZE(sb);
2263 block = le32_to_cpu(gdp[desc].bg_inode_table) +
2264 (offset >> EXT3_BLOCK_SIZE_BITS(sb));
2266 iloc->block_group = block_group;
2267 iloc->offset = offset & (EXT3_BLOCK_SIZE(sb) - 1);
2272 * ext3_get_inode_loc returns with an extra refcount against the inode's
2273 * underlying buffer_head on success. If `in_mem' is false then we're purely
2274 * trying to determine the inode's location on-disk and no read need be
2277 static int ext3_get_inode_loc(struct inode *inode,
2278 struct ext3_iloc *iloc, int in_mem)
2280 unsigned long block;
2281 struct buffer_head *bh;
2283 block = ext3_get_inode_block(inode->i_sb, inode->i_ino, iloc);
2287 bh = sb_getblk(inode->i_sb, block);
2289 ext3_error (inode->i_sb, "ext3_get_inode_loc",
2290 "unable to read inode block - "
2291 "inode=%lu, block=%lu", inode->i_ino, block);
2294 if (!buffer_uptodate(bh)) {
2296 if (buffer_uptodate(bh)) {
2297 /* someone brought it uptodate while we waited */
2302 /* we can't skip I/O if inode is on a disk only */
2304 struct buffer_head *bitmap_bh;
2305 struct ext3_group_desc *desc;
2306 int inodes_per_buffer;
2307 int inode_offset, i;
2312 * If this is the only valid inode in the block we
2313 * need not read the block.
2315 block_group = (inode->i_ino - 1) /
2316 EXT3_INODES_PER_GROUP(inode->i_sb);
2317 inodes_per_buffer = bh->b_size /
2318 EXT3_INODE_SIZE(inode->i_sb);
2319 inode_offset = ((inode->i_ino - 1) %
2320 EXT3_INODES_PER_GROUP(inode->i_sb));
2321 start = inode_offset & ~(inodes_per_buffer - 1);
2323 /* Is the inode bitmap in cache? */
2324 desc = ext3_get_group_desc(inode->i_sb,
2329 bitmap_bh = sb_getblk(inode->i_sb,
2330 le32_to_cpu(desc->bg_inode_bitmap));
2335 * If the inode bitmap isn't in cache then the
2336 * optimisation may end up performing two reads instead
2337 * of one, so skip it.
2339 if (!buffer_uptodate(bitmap_bh)) {
2343 for (i = start; i < start + inodes_per_buffer; i++) {
2344 if (i == inode_offset)
2346 if (ext3_test_bit(i, bitmap_bh->b_data))
2350 if (i == start + inodes_per_buffer) {
2351 /* all other inodes are free, so skip I/O */
2352 memset(bh->b_data, 0, bh->b_size);
2353 set_buffer_uptodate(bh);
2361 * There are another valid inodes in the buffer so we must
2362 * read the block from disk
2365 bh->b_end_io = end_buffer_read_sync;
2366 submit_bh(READ, bh);
2368 if (!buffer_uptodate(bh)) {
2369 ext3_error(inode->i_sb, "ext3_get_inode_loc",
2370 "unable to read inode block - "
2371 "inode=%lu, block=%lu",
2372 inode->i_ino, block);
2382 void ext3_truncate(struct inode * inode)
2384 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
2386 ext3_truncate_nocheck(inode);
2389 void ext3_set_inode_flags(struct inode *inode)
2391 unsigned int flags = EXT3_I(inode)->i_flags;
2393 inode->i_flags &= ~(S_SYNC|S_APPEND|S_IMMUTABLE|S_IUNLINK|S_BARRIER|S_NOATIME|S_DIRSYNC);
2394 if (flags & EXT3_SYNC_FL)
2395 inode->i_flags |= S_SYNC;
2396 if (flags & EXT3_APPEND_FL)
2397 inode->i_flags |= S_APPEND;
2398 if (flags & EXT3_IMMUTABLE_FL)
2399 inode->i_flags |= S_IMMUTABLE;
2400 if (flags & EXT3_IUNLINK_FL)
2401 inode->i_flags |= S_IUNLINK;
2402 if (flags & EXT3_BARRIER_FL)
2403 inode->i_flags |= S_BARRIER;
2404 if (flags & EXT3_NOATIME_FL)
2405 inode->i_flags |= S_NOATIME;
2406 if (flags & EXT3_DIRSYNC_FL)
2407 inode->i_flags |= S_DIRSYNC;
2410 void ext3_read_inode(struct inode * inode)
2412 struct ext3_iloc iloc;
2413 struct ext3_inode *raw_inode;
2414 struct ext3_inode_info *ei = EXT3_I(inode);
2415 struct buffer_head *bh;
2420 #ifdef CONFIG_EXT3_FS_POSIX_ACL
2421 ei->i_acl = EXT3_ACL_NOT_CACHED;
2422 ei->i_default_acl = EXT3_ACL_NOT_CACHED;
2424 ei->i_rsv_window.rsv_end = EXT3_RESERVE_WINDOW_NOT_ALLOCATED;
2426 if (ext3_get_inode_loc(inode, &iloc, 0))
2429 raw_inode = ext3_raw_inode(&iloc);
2430 inode->i_mode = le16_to_cpu(raw_inode->i_mode);
2431 uid = (uid_t)le16_to_cpu(raw_inode->i_uid_low);
2432 gid = (gid_t)le16_to_cpu(raw_inode->i_gid_low);
2433 if(!(test_opt (inode->i_sb, NO_UID32))) {
2434 uid |= le16_to_cpu(raw_inode->i_uid_high) << 16;
2435 gid |= le16_to_cpu(raw_inode->i_gid_high) << 16;
2437 inode->i_uid = INOXID_UID(XID_TAG(inode), uid, gid);
2438 inode->i_gid = INOXID_GID(XID_TAG(inode), uid, gid);
2439 inode->i_xid = INOXID_XID(XID_TAG(inode), uid, gid,
2440 le16_to_cpu(raw_inode->i_raw_xid));
2442 inode->i_nlink = le16_to_cpu(raw_inode->i_links_count);
2443 inode->i_size = le32_to_cpu(raw_inode->i_size);
2444 inode->i_atime.tv_sec = le32_to_cpu(raw_inode->i_atime);
2445 inode->i_ctime.tv_sec = le32_to_cpu(raw_inode->i_ctime);
2446 inode->i_mtime.tv_sec = le32_to_cpu(raw_inode->i_mtime);
2447 inode->i_atime.tv_nsec = inode->i_ctime.tv_nsec = inode->i_mtime.tv_nsec = 0;
2450 ei->i_next_alloc_block = 0;
2451 ei->i_next_alloc_goal = 0;
2452 ei->i_dir_start_lookup = 0;
2453 ei->i_dtime = le32_to_cpu(raw_inode->i_dtime);
2454 /* We now have enough fields to check if the inode was active or not.
2455 * This is needed because nfsd might try to access dead inodes
2456 * the test is that same one that e2fsck uses
2457 * NeilBrown 1999oct15
2459 if (inode->i_nlink == 0) {
2460 if (inode->i_mode == 0 ||
2461 !(EXT3_SB(inode->i_sb)->s_mount_state & EXT3_ORPHAN_FS)) {
2462 /* this inode is deleted */
2466 /* The only unlinked inodes we let through here have
2467 * valid i_mode and are being read by the orphan
2468 * recovery code: that's fine, we're about to complete
2469 * the process of deleting those. */
2471 inode->i_blksize = PAGE_SIZE; /* This is the optimal IO size
2472 * (for stat), not the fs block
2474 inode->i_blocks = le32_to_cpu(raw_inode->i_blocks);
2475 ei->i_flags = le32_to_cpu(raw_inode->i_flags);
2476 #ifdef EXT3_FRAGMENTS
2477 ei->i_faddr = le32_to_cpu(raw_inode->i_faddr);
2478 ei->i_frag_no = raw_inode->i_frag;
2479 ei->i_frag_size = raw_inode->i_fsize;
2481 ei->i_file_acl = le32_to_cpu(raw_inode->i_file_acl);
2482 if (!S_ISREG(inode->i_mode)) {
2483 ei->i_dir_acl = le32_to_cpu(raw_inode->i_dir_acl);
2486 ((__u64)le32_to_cpu(raw_inode->i_size_high)) << 32;
2488 ei->i_disksize = inode->i_size;
2489 inode->i_generation = le32_to_cpu(raw_inode->i_generation);
2490 ei->i_block_group = iloc.block_group;
2491 ei->i_rsv_window.rsv_start = 0;
2492 ei->i_rsv_window.rsv_end= 0;
2493 atomic_set(&ei->i_rsv_window.rsv_goal_size, EXT3_DEFAULT_RESERVE_BLOCKS);
2494 seqlock_init(&ei->i_rsv_window.rsv_seqlock);
2496 * NOTE! The in-memory inode i_data array is in little-endian order
2497 * even on big-endian machines: we do NOT byteswap the block numbers!
2499 for (block = 0; block < EXT3_N_BLOCKS; block++)
2500 ei->i_data[block] = raw_inode->i_block[block];
2501 INIT_LIST_HEAD(&ei->i_orphan);
2503 if (S_ISREG(inode->i_mode)) {
2504 inode->i_op = &ext3_file_inode_operations;
2505 inode->i_fop = &ext3_file_operations;
2506 ext3_set_aops(inode);
2507 } else if (S_ISDIR(inode->i_mode)) {
2508 inode->i_op = &ext3_dir_inode_operations;
2509 inode->i_fop = &ext3_dir_operations;
2510 } else if (S_ISLNK(inode->i_mode)) {
2511 if (ext3_inode_is_fast_symlink(inode))
2512 inode->i_op = &ext3_fast_symlink_inode_operations;
2514 inode->i_op = &ext3_symlink_inode_operations;
2515 ext3_set_aops(inode);
2518 inode->i_op = &ext3_special_inode_operations;
2519 if (raw_inode->i_block[0])
2520 init_special_inode(inode, inode->i_mode,
2521 old_decode_dev(le32_to_cpu(raw_inode->i_block[0])));
2523 init_special_inode(inode, inode->i_mode,
2524 new_decode_dev(le32_to_cpu(raw_inode->i_block[1])));
2527 ext3_set_inode_flags(inode);
2531 make_bad_inode(inode);
2536 * Post the struct inode info into an on-disk inode location in the
2537 * buffer-cache. This gobbles the caller's reference to the
2538 * buffer_head in the inode location struct.
2540 * The caller must have write access to iloc->bh.
2542 static int ext3_do_update_inode(handle_t *handle,
2543 struct inode *inode,
2544 struct ext3_iloc *iloc)
2546 struct ext3_inode *raw_inode = ext3_raw_inode(iloc);
2547 struct ext3_inode_info *ei = EXT3_I(inode);
2548 struct buffer_head *bh = iloc->bh;
2549 uid_t uid = XIDINO_UID(XID_TAG(inode), inode->i_uid, inode->i_xid);
2550 gid_t gid = XIDINO_GID(XID_TAG(inode), inode->i_gid, inode->i_xid);
2551 int err = 0, rc, block;
2553 /* For fields not not tracking in the in-memory inode,
2554 * initialise them to zero for new inodes. */
2555 if (ei->i_state & EXT3_STATE_NEW)
2556 memset(raw_inode, 0, EXT3_SB(inode->i_sb)->s_inode_size);
2558 raw_inode->i_mode = cpu_to_le16(inode->i_mode);
2559 if(!(test_opt(inode->i_sb, NO_UID32))) {
2560 raw_inode->i_uid_low = cpu_to_le16(low_16_bits(uid));
2561 raw_inode->i_gid_low = cpu_to_le16(low_16_bits(gid));
2563 * Fix up interoperability with old kernels. Otherwise, old inodes get
2564 * re-used with the upper 16 bits of the uid/gid intact
2567 raw_inode->i_uid_high =
2568 cpu_to_le16(high_16_bits(uid));
2569 raw_inode->i_gid_high =
2570 cpu_to_le16(high_16_bits(gid));
2572 raw_inode->i_uid_high = 0;
2573 raw_inode->i_gid_high = 0;
2576 raw_inode->i_uid_low =
2577 cpu_to_le16(fs_high2lowuid(uid));
2578 raw_inode->i_gid_low =
2579 cpu_to_le16(fs_high2lowgid(gid));
2580 raw_inode->i_uid_high = 0;
2581 raw_inode->i_gid_high = 0;
2583 #ifdef CONFIG_INOXID_GID32
2584 raw_inode->i_raw_xid = cpu_to_le16(inode->i_xid);
2586 raw_inode->i_links_count = cpu_to_le16(inode->i_nlink);
2587 raw_inode->i_size = cpu_to_le32(ei->i_disksize);
2588 raw_inode->i_atime = cpu_to_le32(inode->i_atime.tv_sec);
2589 raw_inode->i_ctime = cpu_to_le32(inode->i_ctime.tv_sec);
2590 raw_inode->i_mtime = cpu_to_le32(inode->i_mtime.tv_sec);
2591 raw_inode->i_blocks = cpu_to_le32(inode->i_blocks);
2592 raw_inode->i_dtime = cpu_to_le32(ei->i_dtime);
2593 raw_inode->i_flags = cpu_to_le32(ei->i_flags);
2594 #ifdef EXT3_FRAGMENTS
2595 raw_inode->i_faddr = cpu_to_le32(ei->i_faddr);
2596 raw_inode->i_frag = ei->i_frag_no;
2597 raw_inode->i_fsize = ei->i_frag_size;
2599 raw_inode->i_file_acl = cpu_to_le32(ei->i_file_acl);
2600 if (!S_ISREG(inode->i_mode)) {
2601 raw_inode->i_dir_acl = cpu_to_le32(ei->i_dir_acl);
2603 raw_inode->i_size_high =
2604 cpu_to_le32(ei->i_disksize >> 32);
2605 if (ei->i_disksize > 0x7fffffffULL) {
2606 struct super_block *sb = inode->i_sb;
2607 if (!EXT3_HAS_RO_COMPAT_FEATURE(sb,
2608 EXT3_FEATURE_RO_COMPAT_LARGE_FILE) ||
2609 EXT3_SB(sb)->s_es->s_rev_level ==
2610 cpu_to_le32(EXT3_GOOD_OLD_REV)) {
2611 /* If this is the first large file
2612 * created, add a flag to the superblock.
2614 err = ext3_journal_get_write_access(handle,
2615 EXT3_SB(sb)->s_sbh);
2618 ext3_update_dynamic_rev(sb);
2619 EXT3_SET_RO_COMPAT_FEATURE(sb,
2620 EXT3_FEATURE_RO_COMPAT_LARGE_FILE);
2623 err = ext3_journal_dirty_metadata(handle,
2624 EXT3_SB(sb)->s_sbh);
2628 raw_inode->i_generation = cpu_to_le32(inode->i_generation);
2629 if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
2630 if (old_valid_dev(inode->i_rdev)) {
2631 raw_inode->i_block[0] =
2632 cpu_to_le32(old_encode_dev(inode->i_rdev));
2633 raw_inode->i_block[1] = 0;
2635 raw_inode->i_block[0] = 0;
2636 raw_inode->i_block[1] =
2637 cpu_to_le32(new_encode_dev(inode->i_rdev));
2638 raw_inode->i_block[2] = 0;
2640 } else for (block = 0; block < EXT3_N_BLOCKS; block++)
2641 raw_inode->i_block[block] = ei->i_data[block];
2643 BUFFER_TRACE(bh, "call ext3_journal_dirty_metadata");
2644 rc = ext3_journal_dirty_metadata(handle, bh);
2647 ei->i_state &= ~EXT3_STATE_NEW;
2651 ext3_std_error(inode->i_sb, err);
2656 * ext3_write_inode()
2658 * We are called from a few places:
2660 * - Within generic_file_write() for O_SYNC files.
2661 * Here, there will be no transaction running. We wait for any running
2662 * trasnaction to commit.
2664 * - Within sys_sync(), kupdate and such.
2665 * We wait on commit, if tol to.
2667 * - Within prune_icache() (PF_MEMALLOC == true)
2668 * Here we simply return. We can't afford to block kswapd on the
2671 * In all cases it is actually safe for us to return without doing anything,
2672 * because the inode has been copied into a raw inode buffer in
2673 * ext3_mark_inode_dirty(). This is a correctness thing for O_SYNC and for
2676 * Note that we are absolutely dependent upon all inode dirtiers doing the
2677 * right thing: they *must* call mark_inode_dirty() after dirtying info in
2678 * which we are interested.
2680 * It would be a bug for them to not do this. The code:
2682 * mark_inode_dirty(inode)
2684 * inode->i_size = expr;
2686 * is in error because a kswapd-driven write_inode() could occur while
2687 * `stuff()' is running, and the new i_size will be lost. Plus the inode
2688 * will no longer be on the superblock's dirty inode list.
2690 int ext3_write_inode(struct inode *inode, int wait)
2692 if (current->flags & PF_MEMALLOC)
2695 if (ext3_journal_current_handle()) {
2696 jbd_debug(0, "called recursively, non-PF_MEMALLOC!\n");
2704 return ext3_force_commit(inode->i_sb);
2707 int ext3_setattr_flags(struct inode *inode, unsigned int flags)
2709 unsigned int oldflags, newflags;
2712 oldflags = EXT3_I(inode)->i_flags;
2713 newflags = oldflags &
2714 ~(EXT3_IMMUTABLE_FL | EXT3_IUNLINK_FL | EXT3_BARRIER_FL);
2715 if (flags & ATTR_FLAG_IMMUTABLE)
2716 newflags |= EXT3_IMMUTABLE_FL;
2717 if (flags & ATTR_FLAG_IUNLINK)
2718 newflags |= EXT3_IUNLINK_FL;
2719 if (flags & ATTR_FLAG_BARRIER)
2720 newflags |= EXT3_BARRIER_FL;
2722 if (oldflags ^ newflags) {
2724 struct ext3_iloc iloc;
2726 handle = ext3_journal_start(inode, 1);
2728 return PTR_ERR(handle);
2731 err = ext3_reserve_inode_write(handle, inode, &iloc);
2735 EXT3_I(inode)->i_flags = newflags;
2736 inode->i_ctime = CURRENT_TIME;
2738 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2740 ext3_journal_stop(handle);
2748 * Called from notify_change.
2750 * We want to trap VFS attempts to truncate the file as soon as
2751 * possible. In particular, we want to make sure that when the VFS
2752 * shrinks i_size, we put the inode on the orphan list and modify
2753 * i_disksize immediately, so that during the subsequent flushing of
2754 * dirty pages and freeing of disk blocks, we can guarantee that any
2755 * commit will leave the blocks being flushed in an unused state on
2756 * disk. (On recovery, the inode will get truncated and the blocks will
2757 * be freed, so we have a strong guarantee that no future commit will
2758 * leave these blocks visible to the user.)
2760 * Called with inode->sem down.
2762 int ext3_setattr(struct dentry *dentry, struct iattr *attr)
2764 struct inode *inode = dentry->d_inode;
2766 const unsigned int ia_valid = attr->ia_valid;
2768 error = inode_change_ok(inode, attr);
2772 if ((ia_valid & ATTR_UID && attr->ia_uid != inode->i_uid) ||
2773 (ia_valid & ATTR_GID && attr->ia_gid != inode->i_gid) ||
2774 (ia_valid & ATTR_XID && attr->ia_xid != inode->i_xid)) {
2777 /* (user+group)*(old+new) structure, inode write (sb,
2778 * inode block, ? - but truncate inode update has it) */
2779 handle = ext3_journal_start(inode, 4*EXT3_QUOTA_INIT_BLOCKS+3);
2780 if (IS_ERR(handle)) {
2781 error = PTR_ERR(handle);
2784 error = DQUOT_TRANSFER(inode, attr) ? -EDQUOT : 0;
2786 ext3_journal_stop(handle);
2789 /* Update corresponding info in inode so that everything is in
2790 * one transaction */
2791 if (attr->ia_valid & ATTR_UID)
2792 inode->i_uid = attr->ia_uid;
2793 if (attr->ia_valid & ATTR_GID)
2794 inode->i_gid = attr->ia_gid;
2795 if ((attr->ia_valid & ATTR_XID)
2797 && (inode->i_sb->s_flags & MS_TAGXID))
2798 inode->i_xid = attr->ia_xid;
2799 error = ext3_mark_inode_dirty(handle, inode);
2800 ext3_journal_stop(handle);
2803 if (S_ISREG(inode->i_mode) &&
2804 attr->ia_valid & ATTR_SIZE && attr->ia_size < inode->i_size) {
2807 handle = ext3_journal_start(inode, 3);
2808 if (IS_ERR(handle)) {
2809 error = PTR_ERR(handle);
2813 error = ext3_orphan_add(handle, inode);
2814 EXT3_I(inode)->i_disksize = attr->ia_size;
2815 rc = ext3_mark_inode_dirty(handle, inode);
2818 ext3_journal_stop(handle);
2821 if (ia_valid & ATTR_ATTR_FLAG) {
2822 rc = ext3_setattr_flags(inode, attr->ia_attr_flags);
2827 rc = inode_setattr(inode, attr);
2829 /* If inode_setattr's call to ext3_truncate failed to get a
2830 * transaction handle at all, we need to clean up the in-core
2831 * orphan list manually. */
2833 ext3_orphan_del(NULL, inode);
2835 if (!rc && (ia_valid & ATTR_MODE))
2836 rc = ext3_acl_chmod(inode);
2839 ext3_std_error(inode->i_sb, error);
2847 * akpm: how many blocks doth make a writepage()?
2849 * With N blocks per page, it may be:
2854 * N+5 bitmap blocks (from the above)
2855 * N+5 group descriptor summary blocks
2858 * 2 * EXT3_SINGLEDATA_TRANS_BLOCKS for the quote files
2860 * 3 * (N + 5) + 2 + 2 * EXT3_SINGLEDATA_TRANS_BLOCKS
2862 * With ordered or writeback data it's the same, less the N data blocks.
2864 * If the inode's direct blocks can hold an integral number of pages then a
2865 * page cannot straddle two indirect blocks, and we can only touch one indirect
2866 * and dindirect block, and the "5" above becomes "3".
2868 * This still overestimates under most circumstances. If we were to pass the
2869 * start and end offsets in here as well we could do block_to_path() on each
2870 * block and work out the exact number of indirects which are touched. Pah.
2873 int ext3_writepage_trans_blocks(struct inode *inode)
2875 int bpp = ext3_journal_blocks_per_page(inode);
2876 int indirects = (EXT3_NDIR_BLOCKS % bpp) ? 5 : 3;
2879 if (ext3_should_journal_data(inode))
2880 ret = 3 * (bpp + indirects) + 2;
2882 ret = 2 * (bpp + indirects) + 2;
2885 /* We know that structure was already allocated during DQUOT_INIT so
2886 * we will be updating only the data blocks + inodes */
2887 ret += 2*EXT3_QUOTA_TRANS_BLOCKS;
2894 * The caller must have previously called ext3_reserve_inode_write().
2895 * Give this, we know that the caller already has write access to iloc->bh.
2897 int ext3_mark_iloc_dirty(handle_t *handle,
2898 struct inode *inode, struct ext3_iloc *iloc)
2902 /* the do_update_inode consumes one bh->b_count */
2905 /* ext3_do_update_inode() does journal_dirty_metadata */
2906 err = ext3_do_update_inode(handle, inode, iloc);
2912 * On success, We end up with an outstanding reference count against
2913 * iloc->bh. This _must_ be cleaned up later.
2917 ext3_reserve_inode_write(handle_t *handle, struct inode *inode,
2918 struct ext3_iloc *iloc)
2922 err = ext3_get_inode_loc(inode, iloc, 1);
2924 BUFFER_TRACE(iloc->bh, "get_write_access");
2925 err = ext3_journal_get_write_access(handle, iloc->bh);
2932 ext3_std_error(inode->i_sb, err);
2937 * akpm: What we do here is to mark the in-core inode as clean
2938 * with respect to inode dirtiness (it may still be data-dirty).
2939 * This means that the in-core inode may be reaped by prune_icache
2940 * without having to perform any I/O. This is a very good thing,
2941 * because *any* task may call prune_icache - even ones which
2942 * have a transaction open against a different journal.
2944 * Is this cheating? Not really. Sure, we haven't written the
2945 * inode out, but prune_icache isn't a user-visible syncing function.
2946 * Whenever the user wants stuff synced (sys_sync, sys_msync, sys_fsync)
2947 * we start and wait on commits.
2949 * Is this efficient/effective? Well, we're being nice to the system
2950 * by cleaning up our inodes proactively so they can be reaped
2951 * without I/O. But we are potentially leaving up to five seconds'
2952 * worth of inodes floating about which prune_icache wants us to
2953 * write out. One way to fix that would be to get prune_icache()
2954 * to do a write_super() to free up some memory. It has the desired
2957 int ext3_mark_inode_dirty(handle_t *handle, struct inode *inode)
2959 struct ext3_iloc iloc;
2963 err = ext3_reserve_inode_write(handle, inode, &iloc);
2965 err = ext3_mark_iloc_dirty(handle, inode, &iloc);
2970 * akpm: ext3_dirty_inode() is called from __mark_inode_dirty()
2972 * We're really interested in the case where a file is being extended.
2973 * i_size has been changed by generic_commit_write() and we thus need
2974 * to include the updated inode in the current transaction.
2976 * Also, DQUOT_ALLOC_SPACE() will always dirty the inode when blocks
2977 * are allocated to the file.
2979 * If the inode is marked synchronous, we don't honour that here - doing
2980 * so would cause a commit on atime updates, which we don't bother doing.
2981 * We handle synchronous inodes at the highest possible level.
2983 void ext3_dirty_inode(struct inode *inode)
2985 handle_t *current_handle = ext3_journal_current_handle();
2988 handle = ext3_journal_start(inode, 2);
2991 if (current_handle &&
2992 current_handle->h_transaction != handle->h_transaction) {
2993 /* This task has a transaction open against a different fs */
2994 printk(KERN_EMERG "%s: transactions do not match!\n",
2997 jbd_debug(5, "marking dirty. outer handle=%p\n",
2999 ext3_mark_inode_dirty(handle, inode);
3001 ext3_journal_stop(handle);
3008 * Bind an inode's backing buffer_head into this transaction, to prevent
3009 * it from being flushed to disk early. Unlike
3010 * ext3_reserve_inode_write, this leaves behind no bh reference and
3011 * returns no iloc structure, so the caller needs to repeat the iloc
3012 * lookup to mark the inode dirty later.
3015 ext3_pin_inode(handle_t *handle, struct inode *inode)
3017 struct ext3_iloc iloc;
3021 err = ext3_get_inode_loc(inode, &iloc, 1);
3023 BUFFER_TRACE(iloc.bh, "get_write_access");
3024 err = journal_get_write_access(handle, iloc.bh);
3026 err = ext3_journal_dirty_metadata(handle,
3031 ext3_std_error(inode->i_sb, err);
3036 int ext3_change_inode_journal_flag(struct inode *inode, int val)
3043 * We have to be very careful here: changing a data block's
3044 * journaling status dynamically is dangerous. If we write a
3045 * data block to the journal, change the status and then delete
3046 * that block, we risk forgetting to revoke the old log record
3047 * from the journal and so a subsequent replay can corrupt data.
3048 * So, first we make sure that the journal is empty and that
3049 * nobody is changing anything.
3052 journal = EXT3_JOURNAL(inode);
3053 if (is_journal_aborted(journal) || IS_RDONLY(inode))
3056 journal_lock_updates(journal);
3057 journal_flush(journal);
3060 * OK, there are no updates running now, and all cached data is
3061 * synced to disk. We are now in a completely consistent state
3062 * which doesn't have anything in the journal, and we know that
3063 * no filesystem updates are running, so it is safe to modify
3064 * the inode's in-core data-journaling state flag now.
3068 EXT3_I(inode)->i_flags |= EXT3_JOURNAL_DATA_FL;
3070 EXT3_I(inode)->i_flags &= ~EXT3_JOURNAL_DATA_FL;
3071 ext3_set_aops(inode);
3073 journal_unlock_updates(journal);
3075 /* Finally we can mark the inode as dirty. */
3077 handle = ext3_journal_start(inode, 1);
3079 return PTR_ERR(handle);
3081 err = ext3_mark_inode_dirty(handle, inode);
3083 ext3_journal_stop(handle);
3084 ext3_std_error(inode->i_sb, err);